kopt doesn't support model saving for models with custom activation
functions. So in these cases it's not plausible for users to test model
saving with test_fn. Let's make it optional by adding a save_model
argument similar to the one in CompileFN.
Also, users may not wish using test_fn with a test set i.e. a data_fn
that just returns (x_train, y_train), tuple. Right now a second set is mandatory
in data_fn. I think we can relax this by just using the training set in test_fn.
Avsecz
commented
6 years ago
kopt doesn't support model saving for models with custom activation functions. So in these cases it's not plausible for users to test model saving with test_fn. Let's make it optional by adding a save_model argument similar to the one in CompileFN.
Also, users may not wish using test_fn with a test set i.e. a data_fn that just returns
(x_train, y_train),tuple. Right now a second set is mandatory in data_fn. I think we can relax this by just using the training set in test_fn.